CN103218820A - Camera calibration error compensation method based on multi-dimensional characteristics - Google Patents

Abstract

A camera calibration error compensation method based on multi-dimensional characteristics includes the following steps: (1) preparing data: collecting p images of standard targets, obtaining p images with errors, selecting q key points through each images, obtaining p*q key points; (2) extracting the characteristics of the key points: extracting the characteristics of each key point; (3) calculating the actual errors of the p*q key points (delta x, delta y) p*q; (4) conducting simulated training: conducting a support vector regression model training by using support vector machine (SVM) light tools; and (5) estimating errors: obtaining the actual position (x, y) q of q key points, then extracting the characteristics of the q key points according to step (2), storing the characteristics of the q key points in a to-be-regressed characteristic file, and calculating the compensation value (delta x, delta y) of each key point. According to the camera calibration error compensation method based on the multi-dimensional characteristics, association characteristics of a scene image are used, the compensation value of each collected image estimated in support vector regression is adopted, and by means of the camera calibration error compensation method based on the multi-dimensional characteristics, compensated light target center is close to an ideal light target center.

Description

A kind of camera calibration error compensating method based on the multidimensional feature

Technical field

The invention belongs to image processing field, be specifically related to a kind of camera calibration error compensating method based on the multidimensional feature.

Background technology

Camera calibration is one of major issue in the machine vision.For the pin-hole model camera, calibration process mainly is to find the solution camera intrinsic parameter (Intrinsicparameters) and outer parameter (Extrinsicparameters).Arrive scene coordinate (world coordinate system) conversion by this inside and outside parametric solution object pixel coordinate (image coordinate system).Calibration process can adopt direct linear approach, Tsai method, Zhang Zhengyou method etc.But no matter adopt which kind of scaling method,, make to have error between coordinates of targets that actual computation goes out and the ideal coordinates because camera lens radial distortion, inclination geometry deformation, site environment change.Revise the quality of camera calibration error, not only directly influence the accuracy of changes in coordinates, and the accuracy of the follow-up high-rise image understanding of remote effect.

For the existing more research both at home and abroad of camera calibration compensation of error method.People such as Bukhari utilize pedal line method and homogeneous equation model to estimate radial error from single image.People such as Wang propose distortional point and are on the concentric circles at the radial distortion problem, compensate distortion, the effect preferably that obtains by the center of circle of calculating distortional point.People such as Lucchese have proposed the method for correct radial distortion simultaneously and inclination and distortion, but only compensate with the radial distortion of 5 order polynomials needing at least with binomial.People such as Zhang Jiacheng have mixed multiple compensate for estimated model, with classical model fault image is proofreaied and correct for the first time, carry out the secondary fine correction with the Polyhedral Function Fitting Method method, carrying out gray scale with cubic B-spline function rebuilds, compare single lens distortion calibration model, precision improves, and robustness strengthens, after the correction radially root-mean-square error be 0.3 pixel.People such as Liu Tang have set up the conjunctive model of radial distortion and oblique distortions, try to achieve the distortion parameter of standard grid with least square method and optimization algorithm, at last target are compensated.

Above-mentioned compensation method has all obtained certain effect, but has two aspects still can continue to improve.The first, with high-order moment offset is carried out modeling, along with number of times increases, calculated amount is exponential quick growth, is difficult to the contradiction between EQUILIBRIUM CALCULATION FOR PROCESS time and the compensation quality; The second, camera distortion not only cause departing from of regional area pixel position, also influenced the imaging of view picture figure.So choosing of characteristics of image must take into account local feature and global characteristics.

Camera calibration error compensation model at present commonly used be camera calibration by finding the solution inside and outside parameter, realize that object pixel coordinate (image coordinate system) is to scene coordinate (world coordinate system) conversion.The camera calibration error model is as follows:

Order $A=\left[\begin{array}{ccc}{f}_x& 0& c_x\\ 0& f_y& c_y\\ 0& 0& 1\end{array}\right],$

In the formula, (X, Y Z) are the world coordinates of a point, and (u v) is the pixel coordinate of point on image, and A is the intrinsic parameter of camera, (c _x, c _y) be the principal point position, f _xWith f _yBe camera focus, they are unit with the pixel.Rotation-translation matrix R|T is called outer parameter, and it has described the pose of camera with respect to scene.

Be camera compensation of error function, R is the proper vector (r of current environment ₁, r ₂..., r _n), can be the feature of various sign current environment situations such as average light illumination, average gray, gray variance.So-called compensation is exactly the penalty function that finds an optimum that depends on R

Summary of the invention

Technical matters to be solved by this invention provides a kind of camera calibration compensation of error method based on the multidimensional feature.

For solving above technical matters, the present invention adopts following technical scheme:

A kind of camera calibration compensation of error method based on the multidimensional feature, it may further comprise the steps:

(1) prepares data: gather p width of cloth standard target target image earlier, obtain the image that the p width of cloth has error, from every width of cloth image, choose q key point then, obtain p * q described key point;

(2) feature of extraction key point: extract the feature of each described key point, described feature comprises color characteristic, local Gabor feature and global association feature;

(3) actual error (Δ x, Δ y) of p * q described key point of calculating _{P * q}: the actual error (Δ x, Δ y) of the ideal position coordinate of each key point coordinate and this key point among the calculating p width of cloth figure _{P * q}

(4) simulated training: adopt the SVMLight instrument to carry out the support vector regression model training, with (Δ x, the Δ y) that obtains in the feature of p * q described key point obtaining in the described step (2) and the described step (3) _{P * q}As input, obtain model file at last;

(5) estimation error: when photographing the new picture of a width of cloth, obtain q key point physical location (x, y) _q, (2) extract the feature of this q described key point then set by step, deposit in the tag file that needs to return, and calculate the offset (Δ x, Δ y) of each described key point.

Further, the color characteristic that is extracted in the described step (2) be around each described key point in N * n-quadrant the average of the color component of 4 kinds of color spaces commonly used and variance as feature, described 4 kinds of color spaces commonly used are RGB, CMYK, HSV, HIS, and wherein, N is smaller or equal to 50.

Further again, local Gabor Feature Extraction method may further comprise the steps in the described step (2):

A. coloured image is normalized to gray level and is 256 gray-scale map;

B. be divided into the subwindow of L N * N around the current key point, wherein, L is smaller or equal to 50;

C. each window is carried out the Gabor conversion, original image is F(x, y), and through obtaining new image Q(x after the Gabor conversion, y), wherein,

Q(x,y)=[(Gabor _R(x,y) ^*F(x,y)) ²+(Gabor _I(x,y) ^*F(x,y)) ²] ^1/2 （1）

In the formula (1),

${\mathrm{Gabor}}_R(x,y)=\frac{1}{{2\mathrm{\&pi;\&sigma;}}^2}{e}^{-\frac{x^2+y^2}{2\mathrm{\&sigma;}}}\mathrm{<figure-callout\; id="2"\; label="cos"\; filenames="HDA00003083703900021.png"\; state="\{\{state\}\}">cos</figure-callout>}\frac{2\mathrm{\&pi;}(x\cos \mathrm{\&theta;}+y\sin \mathrm{\&theta;})}{l}---\left(2\right)$

${\mathrm{Gabor}}_I(x,y)=\frac{1}{{2\mathrm{\&pi;\&sigma;}}^2}{e}^{-\frac{x^2+y^2}{2\mathrm{\&sigma;}}}\sin \frac{2\mathrm{\&pi;}(x\cos \mathrm{\&theta;}+y\sin \mathrm{\&theta;})}{l}---\left(3\right)$

σ=π in formula (2), (3);

Behind the computational transformation, the average of each window gray scale is represented that by μ the standard deviation of each window is represented that by δ (k=δ/μ) expression at last, obtains the local Gabor feature of L window: μ to the Z-factor of each window by k ₁, μ ₂..., μ _i..., μ _L, δ ₁, δ ₂..., δ _i..., δ _LWith k ₁, k ₂..., k _i..., k _L, total 3L dimension, wherein, i is a series of windows number.

Further, the global association Feature Extraction may further comprise the steps in the described step (2):

A. can regard local Gabor feature as three one-dimensional sequence: μ ₁, μ ₂..., μ _i..., μ _L, δ ₁, δ ₂..., δ _i..., δ _LWith k ₁, k ₂..., k _i..., k _L, wherein i represents the window sequence number;

B. utilize autocorrelation function compute associations feature, autocorrelation function is suc as formula (4), and wherein, m is relevant exponent number, and L is the number of Gabor subwindow, and m＜L,

$R_{\mathrm{\&mu;n}}=\frac{1}{L-n}\underset{i=1}{\overset{L-n}{\mathrm{\&Sigma;}}}{\mathrm{\&mu;}}_i{\mathrm{\&mu;}}_{i+n},(n=\mathrm{1,2},...,m)---\left(4\right)$

This moment is by sequence μ ₁, μ ₂..., μ _i..., μ _LCan calculate m feature R _{μ 1}, R _{μ 2}..., R _{μ m}, same can calculate R _{δ 1}, R _{δ 2}..., R _{δ m}With R _K1, R _K2..., R _Km, be total to 3m global association feature.

Further, the actual error of the calculating p * q key point in the described step (3) (Δ x, Δ y) _{P * q}Method may further comprise the steps:

A, will from standard target is marked on a map, obtain q key point laser measurement the center as the ideal position coordinate (x, y) _s, wherein, s=1,2 ..., q;

B, with the some recognition technology obtain q key point among the p width of cloth figure physical location (x, y) _{T * s}, wherein, t=1,2 ..., p, s=1,2 ..., q;

C, calculating: (Δ x, Δ y) _{P * q}=| (x, y) _s-(x, y) t * s|.

Further, the SVMLight instrument is adopted in the simulated training in the described step (4), and its step is as follows:

A. import training order, described training order is: svm_learn – zr – t2 – g0.12example_filemodel_file, and wherein, example_file is the input file of training data; Model_file is the model result file of output; It is r that Can is Shuoed – z, the expression Regression Model; It is 2 that Can is Shuoed – t, and the radially basic kernel function of Gauss (RBF) is adopted in expression; It is 0.12 that Can is Shuoed – g, and the expression Gauss radially parameter gamma of base nuclear is 0.12.

B. the generation model file need generate two different model files respectively for x, y.

Further, the step of the estimation error in the described step (5) is as follows:

A. input returns order, described recurrence order is svm_classify example_file model_file output_file, wherein, the tag file of example_file for needing to return, model_file is the model file of the generation in the described step (4), and output_file is an output file;

B. generate output file, each row is corresponding one by one with row in the tag file that need to return in the described output file, and expression needs each regressand value of going in the tag file of recurrence.

Because adopt above technical scheme, the present invention compared with prior art has following advantage:

The characteristics of image that the present invention extracted has been taken into account local feature and global characteristics, and utilize the linked character of image scene, adopt support vector regression to estimate the compensation rate of every width of cloth images acquired in real time, adopt the method for the present invention light target center after the compensation that makes more to approach the desired light pinwheel.

With the method that traditional employing high-order moment is carried out modeling to offset, many feature extracting methods of the present invention, adopt the quantity that increases training image to improve precision of prediction, and the computing time of compensating error and training time are irrelevant.

Description of drawings

Fig. 1 is the standard target picture of marking on a map;

Fig. 2 is 9 subwindow figure;

Fig. 3 is the distribution situation of central point in the different distance scope before and after the compensation;

Fig. 4 is an emulation bridge original image;

Fig. 5 is the contrast of the coordinate after compensating with the coordinate that does not compensate.

Embodiment

Further set forth the present invention below in conjunction with accompanying drawing.

A kind of camera calibration error compensating method based on the multidimensional feature specifically may further comprise the steps:

(1) prepare data: gather p width of cloth standard target target image earlier, the standard target obtains the image that the p width of cloth has error as shown in Figure 1, chooses q key point then from every width of cloth image, obtains p * q key point, and wherein p, q are positive integer;

(2) feature of extraction key point:

Extract the feature of each described key point, each key point has three category features, is respectively color characteristic, local Gabor feature and global association feature, and such key point has p * q.The extracting method of three category features is as follows:

(21) extracting method of color characteristic adopts color characteristic extracting method commonly used, extracts around each key point in N * n-quadrant the average of color component of 4 kinds of color space RGB commonly used, CMYK, HSV, HIS and variance as feature.Wherein, the RGB color space has three color components, and 3 * 2 dimensional features are arranged; The CMYK color space has three color components, and 3 * 2 dimensional features are arranged; There are three color components in the hsv color space, and 3 * 2 dimensional features are arranged, and the HSI color space has four color components, and 4 * 2 dimensional features are arranged, thus have 26 dimension color characteristics, in this example, N=8.

(22) local Gabor Feature Extraction method, concrete steps are as follows:

A. coloured image is normalized to gray level and is 256 gray-scale map;

B. will be divided into the subwindow of L individual 8 * 8 around the current key point M, in this example, L=9, as shown in Figure 2;

C. each window is carried out the Gabor conversion, original image is F(x, y), and through obtaining new image Q(x after the Gabor conversion, y), wherein,

Q(x,y)=[(Gabor _R(x,y) ^*F(x,y)) ²+(Gabor _I(x,y) ^*F(x,y)) ²] ^1/2 （1）

In the formula (1),

${\mathrm{Gabor}}_R(x,y)=\frac{1}{{2\mathrm{\&pi;\&sigma;}}^2}{e}^{-\frac{x^2+{\mathrm{<figure-callout\; id="2"\; label="y"\; filenames="HDA00003083703900021.png"\; state="\{\{state\}\}">y</figure-callout>}}^2}{2\mathrm{\&sigma;}}}\mathrm{<figure-callout\; id="2"\; label="cos"\; filenames="HDA00003083703900021.png"\; state="\{\{state\}\}">cos</figure-callout>}\frac{2\mathrm{\&pi;}(x\cos \mathrm{\&theta;}+y\sin \mathrm{\&theta;})}{l}---\left(2\right)$

${\mathrm{Gabor}}_I(x,y)=\frac{1}{{2\mathrm{\&pi;\&sigma;}}^2}{e}^{-\frac{x^2+{\mathrm{<figure-callout\; id="2"\; label="y"\; filenames="HDA00003083703900021.png"\; state="\{\{state\}\}">y</figure-callout>}}^2}{2\mathrm{\&sigma;}}}\mathrm{<figure-callout\; id="2"\; label="sin"\; filenames="HDA00003083703900021.png"\; state="\{\{state\}\}">sin</figure-callout>}\frac{2\mathrm{\&pi;}(x\cos \mathrm{\&theta;}+y\sin \mathrm{\&theta;})}{l}---\left(3\right)$

σ=π in formula (2), (3),

Behind the computational transformation, the average of each window gray scale is represented that by μ the standard deviation of each window is represented that by δ the Z-factor of each window is by k (k=δ/μ) expression.At last, obtain the local Gabor feature of 9 windows: μ ₁, μ ₂..., μ _i..., μ ₉, δ ₁, δ ₂..., δ _i..., δ _LWith k ₁, k ₂..., k _i..., k ₉, have 27 dimensions, wherein, i is a series of windows number.

(23) global association Feature Extraction method, step is as follows:

A. can regard local Gabor feature as three one-dimensional sequence: μ ₁, μ ₂..., μ _i..., μ ₉, δ ₁, δ ₂..., δ _i..., δ _LWith k ₁, k ₂..., k _i..., k ₉, wherein i represents the window sequence number;

B. utilize autocorrelation function compute associations feature, autocorrelation function is suc as formula (4), and wherein, m is relevant exponent number, and L is the number of Gabor subwindow, and m＜L.

$R_{\mathrm{\&mu;n}}=\frac{1}{L-n}\underset{i=1}{\overset{L-n}{\mathrm{\&Sigma;}}}{\mathrm{\&mu;}}_i{\mathrm{\&mu;}}_{i+n},(n=\mathrm{1,2},...,m)---\left(4\right)$

In this example, m=3, this moment is by sequence μ ₁, μ ₂..., μ _i..., μ ₉Can calculate three feature R _{μ 1}, R _{μ 2}, R _{μ 3}, same can calculate R _{δ 1}, R _{δ 2}, R _{δ 3}With R _K1, R _K2, R _K3, totally 9 global association features.

(3) actual error (Δ x, Δ y) of p * q key point of calculating _{P * q}: the actual error (Δ x, Δ y) of the ideal position coordinate of each key point coordinate and these key points among the calculating p width of cloth figure _{P * q}, its method step is as follows:

A, will from Fig. 1, obtain q key point laser measurement the center as the ideal position coordinate (x, y) _s, wherein, s=1,2 ..., q;

B, with the some recognition technology obtain q key point among the p width of cloth figure physical location (x, y) t * s, wherein, t=1,2 ..., p, s=1,2 ..., q;

C, calculating: (Δ x, Δ y) _{P * q}=| (x, y) s-(x, y) t * s|.

(4) simulated training:

Adopt the SVMLight instrument to carry out support vector regression (SVM) model training, with (Δ x, the Δ y) that obtains in the feature of p * q key point obtaining in the step (2) and the step (3) _{P * q}As input, obtain model file model_file at last, step is as follows:

A. import training order

Training order is: svm_learn – zr – t2 – g0.12example_filemodel_file

Wherein, example_file is the input file of training data; Model_file is the model result file of output; It is r that Can is Shuoed – z, the expression Regression Model; It is 2 that Can is Shuoed – t, and the radially basic kernel function of Gauss (RBF) is adopted in expression; It is 0.12 that Can is Shuoed – g, and the expression Gauss radially parameter gamma of base nuclear is 0.12.

B. form the form following (following only is form type) of example_file file:

Training data of each line display comprises the distance and 62 dimensional features of compensation, totally 62 row, and form of each row is as follows:

Δx1：0.652：0.78…62：0.32

Same, Δ y1:0.652:0.78 ... 62:0.32

C. generation model file: model_file

In this example, need be x, y generates two different model files respectively.

(5) estimation error: when photographing a secondary new picture, the physical location that obtains q key point when photographing the new picture of a pair, obtain q key point physical location (x, y) _q, (2) extract the feature of this q key point then set by step, deposit in the tag file that needs to return, and calculate the offset (Δ x, Δ y) of each key point, and its step is as follows:

A. input returns order

Returning order is: svm_classifyexample_filemodel_fileoutput_file

At this moment, the tag file of example_file for needing to return; Model_file is the model file of the generation in the step (4); Output_file is an output file.

B. the form that forms the example_file file is as follows:

One of each line display returns feature, comprises 62 dimensional features, its form following (following only is form type):

1：0.652：0.78…62：0.32

C. generate the output_file file

In the output_file file, its form is corresponding one by one with the row of example_file for each row, the regressand value of each row among the expression example_file.

(6) emulation experiment

In this experiment, the error compensation that 10 monitoring points on the emulation bridge have been carried out 50 times is tested, as Fig. 4 is emulation bridge original image, 1～10 stain among Fig. 4 is 10 monitoring points, add up respectively in 50 experiments, calculated the distance of position in the desirable light target, light target center before and after the compensation earlier, the distribution situation of central point in the different distance scope before and after the statistic compensation again, distribution situation is as shown in Figure 3.Here we the center of laser measurement as desired light pinwheel position.Abscissa axis label 1 is represented respectively to compensate forward and backward central point distribution with 1_SVR among Fig. 2, and black represents to put apart from desired center the number of 10 pixels; The ash colour specification is apart from the number of desired center point 11 to 20 pixels; Dark grey is represented the number apart from desired center point 21 to 30 pixels; The white expression is apart from the number of desired center point 30 above pixels.

As can be seen from Figure 3, compensation back, No. 1, No. 4, No. 5, No. 7, No. 9 monitoring points is than having more point nearer apart from ideal point before and after the compensation; No. 2, No. 3, No. 6 both performances of monitoring point are suitable; Before slightly being worse than compensation after No. 8, No. 10 compensation.All in all dynamic compensation method of the present invention has played compensating action to the light target center.

Fig. 5 has provided the contrast of coordinate and the coordinate that does not compensate after the compensation, the target spot center after as can be seen from the figure most of compensation than the target spot center of not compensated more near the position of target on the laser measurement bridge.

The foregoing description only is explanation technical conceive of the present invention and characteristics; its purpose is to allow the personage who is familiar with this technology can understand content of the present invention and enforcement according to this; can not limit protection scope of the present invention with this; all equivalences that spirit is done according to the present invention change or modify, and all should be encompassed within protection scope of the present invention.

Claims (7)

1. camera calibration error compensating method based on the multidimensional feature, it is characterized in that: it may further comprise the steps:

(1) prepares data: gather p width of cloth standard target target image earlier, obtain the image that the p width of cloth has error, from every width of cloth image, choose q key point then, obtain p * q described key point;

(2) feature of extraction key point: extract the feature of each described key point, described feature comprises color characteristic, local Gabor feature and global association feature;

(3) actual error (Δ x, Δ y) of p * q described key point of calculating _{P * q}: the actual error (Δ x, Δ y) of the ideal position coordinate of each key point coordinate and this key point among the calculating p width of cloth figure _{P * q}

(4) simulated training: adopt the SVMLight instrument to carry out the support vector regression model training, with (Δ x, the Δ y) that obtains in the feature of p * q described key point obtaining in the described step (2) and the described step (3) _{P * q}As input, obtain model file at last;

(5) estimation error: when photographing the new picture of a width of cloth, obtain q key point physical location (x, y) _q, (2) extract the feature of this q described key point then set by step, deposit in the tag file that needs to return, and calculate the offset (Δ x, Δ y) of each described key point.

2. the camera calibration error compensating method based on the multidimensional feature according to claim 1, it is characterized in that: the color characteristic that is extracted in the described step (2) be around each described key point in N * n-quadrant the average of the color component of 4 kinds of color spaces commonly used and variance as feature, described 4 kinds of color spaces commonly used are RGB, CMYK, HSV, HIS, wherein, N is smaller or equal to 50.

3. the camera calibration error compensating method based on the multidimensional feature according to claim 2 is characterized in that: local Gabor Feature Extraction method may further comprise the steps in the described step (2):

A. coloured image is normalized to gray level and is 256 gray-scale map;

B. be divided into the subwindow of L N * N around the current key point, wherein, L is smaller or equal to 50;

C. each window is carried out the Gabor conversion, original image is F(x, y), and through obtaining new image Q(x after the Gabor conversion, y), wherein,

Q(x,y)=[(Gabor _R(x,y) ^*F(x,y)) ²+(Gabor _I(x,y) ^*F(x,y)) ²] ^1/2 （1）

In the formula (1),

${\mathrm{Gabor}}_R(x,y)=\frac{1}{{2\mathrm{\&pi;\&sigma;}}^2}{e}^{-\frac{x^2+y^2}{2\mathrm{\&sigma;}}}\cos \frac{2\mathrm{\&pi;}(x\cos \mathrm{\&theta;}+y\sin \mathrm{\&theta;})}{l}---\left(2\right)$

${\mathrm{Gabor}}_I(x,y)=\frac{1}{{2\mathrm{\&pi;\&sigma;}}^2}{e}^{-\frac{x^2+y^2}{2\mathrm{\&sigma;}}}\sin \frac{2\mathrm{\&pi;}(x\cos \mathrm{\&theta;}+y\sin \mathrm{\&theta;})}{l}---\left(3\right)$

σ=π in formula (2), (3);

Behind the computational transformation, the average of each window gray scale is represented that by μ the standard deviation of each window is represented that by δ (k=δ/μ) expression at last, obtains the local Gabor feature of L window: μ to the Z-factor of each window by k ₁, μ ₂..., μ _i..., μ _L, δ ₁, δ ₂..., δ _i..., δ _LWith k ₁, k ₂..., k _i..., k _L, total 3L dimension, wherein, i is a series of windows number.

4. the camera calibration error compensating method based on the multidimensional feature according to claim 3 is characterized in that: the global association Feature Extraction may further comprise the steps in the described step (2):

A. can regard local Gabor feature as three one-dimensional sequence: μ ₁, μ ₂..., μ _i..., μ _L, δ ₁, δ ₂..., δ _i..., δ _LWith k ₁, k ₂..., k _i..., k _L, wherein i represents the window sequence number;

B. utilize autocorrelation function compute associations feature, autocorrelation function is suc as formula (4), and wherein, m is relevant exponent number, and L is the number of Gabor subwindow, and m＜L,

$R_{\mathrm{\&mu;n}}=\frac{1}{L-n}\underset{i=1}{\overset{L-n}{\mathrm{\&Sigma;}}}{\mathrm{\&mu;}}_i{\mathrm{\&mu;}}_{i+n},(n=\mathrm{1,2},...,m)---\left(4\right)$

This moment is by sequence μ ₁, μ ₂..., μ _i..., μ _LCan calculate m feature R _{μ 1}, R _{μ 2}..., R _{μ m}, same can calculate R _{δ 1}, R _{δ 2}..., R _{δ m}With R _K1, R _K2..., R _Km, be total to 3m global association feature.

5. the camera calibration error compensating method based on the multidimensional feature according to claim 1 is characterized in that: the actual error of the calculating p * q key point in the described step (3) (Δ x, Δ y) _{P * q}Method may further comprise the steps:

A, will from standard target is marked on a map, obtain q key point laser measurement the center as the ideal position coordinate (x, y) _s, wherein, s=1,2 ..., q;

B, with the some recognition technology obtain q key point among the p width of cloth figure physical location (x, y) _{T * s}, wherein, t=1,2 ..., p, s=1,2 ..., q;

C, calculating: (Δ x, Δ y) _{P * q}=| (x, y) _s-(x, y) _{T * s}|.

6. the camera calibration error compensating method based on the multidimensional feature according to claim 1 is characterized in that: the SVMLight instrument is adopted in the simulated training in the described step (4), and its step is as follows:

A. import training order, described training order is: svm_learn – zr – t2 – g0.12example_filemodel_file, and wherein, example_file is the input file of training data; Model_file is the model result file of output; It is r that Can is Shuoed – z, the expression Regression Model; It is 2 that Can is Shuoed – t, and the radially basic kernel function of Gauss (RBF) is adopted in expression; It is 0.12 that Can is Shuoed – g, and the expression Gauss radially parameter gamma of base nuclear is 0.12;

B. the generation model file need generate two different model files respectively for x, y.

7. the camera calibration error compensating method based on the multidimensional feature according to claim 1 is characterized in that: the step of the estimation error in the described step (5) is as follows:

A. input returns order, described recurrence order is svm_classify example_file model_file output_file, wherein, the tag file of example_file for needing to return, model_file is the model file of the generation in the described step (4), and output_file is an output file;

B. generate output file, each row is corresponding one by one with row in the tag file that need to return in the described output file, and expression needs each regressand value of going in the tag file of recurrence.

Images